Cluster C · State strategies

Vernacular AI for State Elections: Rajasthan, Bihar, Maharashtra

How vernacular AI changes the math in three crucial Indian state elections — Marwari/Mewari for Rajasthan, Bhojpuri/Maithili for Bihar, Marathi dialects for Maharashtra. Operational deep dive.

7 min readUpdated 22 May 20261,382 words

Three states crucial to Indian electoral politics — Rajasthan, Bihar, Maharashtra — each have language conditions that make standard-Hindi or English-only AI agents almost useless. Each requires its own vernacular configuration. Each rewards campaigns that get the dialect right disproportionately, and punishes campaigns that don't.

This guide is the state-by-state operational deep dive. The dialect map, the engagement data, the system-prompt patterns and the specific moves that work in each state.

Rajasthan

Population: ~8 crore. Vidhan Sabha seats: 200. Last election: November 2023. Next election: 2028.

The dialect map

Rajasthan's language conditions are unusually layered.

Marwari belt (Western Rajasthan). Jodhpur, Bikaner, Jaisalmer, Barmer, Pali, Nagaur, Jalore, Sirohi. Marwari is the primary mother tongue for 80%+ of voters in this region. Standard Hindi is understood but feels formal/foreign.

Mewari belt (Southern Rajasthan). Udaipur, Chittorgarh, Rajsamand, Banswara, Dungarpur. Mewari dominates rural areas; Hindi mixes in urban centres.

Dhundhari belt (Eastern Rajasthan). Jaipur, Ajmer, Tonk, Bundi, Bharatpur (partly). Dhundhari is the local Hindi register — closer to Khari Boli with some unique vocabulary.

Shekhawati belt (North-East Rajasthan). Sikar, Jhunjhunu, Churu. Shekhawati dialect close to Marwari but with distinct features.

Hadoti / Brij belt (South-East). Kota, Bundi, Jhalawar. Hadoti dialect with Braj Bhasha influences.

Plus: Urdu in pockets (Tonk, parts of Jaipur Old City), Gujarati in border districts.

What works in Rajasthan

System prompt pattern (Hindi + Marwari):

यदि caller Marwari में बात करे:
- थारो/म्हारो pronouns
- "जासी/होवेगा/करसी" verb forms
- "काका/काकी/भाईसा/बहनजी" honorifics
- "अरे, ध्यान दीजै/मतलब बताओ ने" जैसे conversational fillers

Sample conversational opening:

"नमस्कार, मैं [Agent name] हूँ, [Candidate]'s AI सहायक। म्हारो साहब बीकानेर के हर परिवार से बात करणो चाहै हैं। बताइए, थारे इलाके की सबसे बड़ी समस्या क्या है?"

Note the natural code-mixing — "बात करणो चाहै हैं" (Marwari verb construction) inside an otherwise standard Hindi sentence. This is the texture that signals authenticity.

Engagement data points

From state pilot deployments in 2024-25:

  • Hindi-only agent: 32-38% completion rate in Marwari belt
  • Marwari-tuned agent: 58-68% completion rate same region
  • Same agent with Mewari tuning: 55-62% in Mewari belt
  • Standard Hindi-with-Mewari-mixin: still ~40% — partial tuning is not enough

Specific issues that resonate

Manifesto + sentiment data from Rajasthan voters tends to surface:

  • Water scarcity (state-defining issue, varies by district)
  • Mukhyamantri Ayushman Arogya Yojana (MAA-Y) and other health schemes
  • Bhamashah / Jan Aadhaar card-related queries
  • Female safety in specific districts (Jaipur, Kota)
  • Migration to Gujarat/Mumbai/Bangalore for work
  • Rural electrification gaps

Bihar

Population: ~13 crore. Vidhan Sabha seats: 243. Last election: November 2025. Next election: 2030.

The dialect map

Bihar's language conditions are even more fragmented than Rajasthan's.

Bhojpuri belt (Western Bihar). Champaran (East + West), Chapra, Siwan, Gopalganj, Saran, Bhojpur, Buxar. Bhojpuri is primary mother tongue. Cross-border with Eastern UP.

Maithili belt (North-Eastern Bihar). Madhubani, Darbhanga, Samastipur, Begusarai, Sitamarhi, Sheohar. Maithili has Sanskrit and Bengali influences; recognised as a scheduled language.

Magahi belt (Southern Bihar). Patna (partly), Gaya, Nalanda, Aurangabad, Nawada, Jehanabad. Magahi has distinct vocabulary and grammar.

Angika belt (Eastern Bihar). Bhagalpur, Banka, Munger, Khagaria. Angika is a distinct dialect with influences from Bengali and Maithili.

Urban Hindi. Patna, parts of major district headquarters. Standard Hindi dominates.

What works in Bihar

For each dialect, distinct system prompt configurations. Sample for Bhojpuri:

यदि caller Bhojpuri में बात करे:
- "रउआ/हम" pronouns
- "जाइब/करब/होई" verb forms
- "भईया/दीदी/चाचा" संबोधन
- "देखीं/सुनीं/बताईं" imperative forms

For Maithili:

यदि caller Maithili में बात करे:
- "अहाँ/हम" pronouns
- "जाएब/करब/होएब" verb forms
- "भाइ/बहिनि/काका" honorifics
- "देखी/सुनी" imperative forms

Engagement data points

From the November 2025 cycle:

  • Hindi-only: 28-35% completion in Bhojpuri belt; 25-30% in Maithili belt
  • Bhojpuri-tuned: 60-70% in Bhojpuri belt
  • Maithili-tuned: 55-65% in Maithili belt
  • Magahi-tuned: 50-58% in Magahi belt (Magahi has less commercial TTS support, so quality is slightly behind)

Note: Bihar's dialect engagement uplift is larger than Rajasthan's. The dialect identity is stronger.

Specific issues that resonate

  • Migration (Bihar is the largest source of out-migration in India)
  • Education access and government school quality
  • Agricultural pricing (rice, wheat MSP)
  • Caste-specific schemes (MBC, EBC, Mahadalit categorisations)
  • Law and order in specific districts
  • Patna metro / connectivity infrastructure

Maharashtra

Population: ~13 crore. Vidhan Sabha seats: 288. Last election: November 2024. Next election: 2029.

The dialect map

Marathi has fewer truly distinct dialects than Hindi but meaningful regional registers.

Standard Marathi (Western Maharashtra). Pune, Satara, Sangli, Kolhapur. The textbook Marathi.

Mumbai/Konkan Marathi (Coastal). Mumbai, Thane, Raigad, Ratnagiri, Sindhudurg. Strong Konkani influences; some Portuguese-era loanwords in coastal districts.

Vidarbha Marathi (Eastern Maharashtra). Nagpur, Amravati, Akola, Yavatmal, Wardha, Chandrapur. Strong Hindi influences in vocabulary; Telugu border influences in southern districts.

Marathwada Marathi (Central-East). Aurangabad, Beed, Latur, Nanded, Osmanabad. Historical Urdu/Persian influences from Nizam era.

Khandesh Marathi (North-West). Jalgaon, Dhule, Nandurbar. Some Gujarati and Hindi influences.

What works in Maharashtra

Standard Marathi works in urban Pune/Mumbai/Aurangabad. Regional registers are critical for rural deployments.

Sample system prompt for Vidarbha:

Vidarbha के काह विशेष:
- Hindi-Marathi mix natural — voter may use either freely
- "होय/नाही/आहे" affirmatives
- "तू" instead of "तुम्ही" in informal contexts
- "बेस/छान" (good) and "बंदा/मायबाप" addressing patterns specific to region

Sample opening for a candidate in Nagpur:

"नमस्कार, मी [Agent name] आहे, [Candidate]'s AI साथीदार. आपण कसे आहात? आम्हाला आपल्या इलाख्यातील सर्वात मोठ्या समस्या जाणून घ्यायच्या आहेत. सांगा, काय परिस्थिती आहे?"

Engagement data points

From the November 2024 cycle:

  • Standard Marathi in Pune urban: 55-62% completion
  • Standard Marathi in Vidarbha rural: 40-45% (the standard register feels too "Pune" to rural voters)
  • Vidarbha-tuned: 58-65%
  • Marathwada-tuned: 55-60%
  • Hindi-only: 35-40% in non-urban areas (despite Maharashtra being a Hindi-speaking state for most voters)

Specific issues that resonate

  • Agricultural distress, especially in Vidarbha (cotton farmer suicides)
  • Urban-rural divide (Mumbai/Pune vs the rest of the state)
  • Water disputes (Krishna, Godavari)
  • Reservation politics (Maratha, OBC)
  • Industrial revival in Marathwada
  • Mumbai-Pune connectivity infrastructure

Cross-state operational patterns

Three patterns that emerge across all three states.

1. Configure once, deploy many

The system prompt template is parameterised. Each state has its own variant. Adding a new district within a state typically requires only minor tuning (additional dialect examples, region-specific issue list).

2. Native-speaker QA is non-negotiable

For each new dialect configuration, get 30-50 native speakers from that exact region to call the agent and rate. Anything under 4/5 on "feels natural" means more tuning before launch. This step adds 2-3 weeks but cannot be skipped.

3. State-level compliance officer

ECI compliance is interpreted slightly differently by state CEOs. A state-level compliance officer who works with the local CEO office reduces friction significantly.

What stays the same across states

Despite the dialect differences, some patterns are universal:

  • AI must self-disclose in the opening line
  • HARD STOP rules (goodbye, anger, silence) apply identically
  • DLT-template registration is required identically
  • DPDP residency is required identically
  • The 12 use cases (see the use cases guide) work in all three states with minor tuning

What this means for 2027-2029 cycles

Looking ahead:

  • Rajasthan 2028: A major opportunity for vernacular-first AI. Marwari and Mewari tuning will be decisive.
  • Bihar 2030: Bhojpuri / Maithili / Magahi / Angika multi-config will set the bar. Campaigns that ship Hindi-only will be punished.
  • Maharashtra 2029: The Vidarbha and Marathwada tuning will determine whether a campaign reaches rural Maharashtra. Mumbai/Pune-only Marathi is insufficient.

Where AiSewak fits

AiSewak's pre-tuned dialect configurations:

  • Rajasthan: Marwari, Mewari, Dhundhari, Shekhawati, Hadoti
  • Bihar: Bhojpuri, Maithili, Magahi, Angika
  • Maharashtra: Standard Marathi, Vidarbha Marathi, Marathwada Marathi, Konkan Marathi, Khandeshi

Plus the universal layers (Hindi, English, code-switching). Adding a new dialect typically takes 5-7 working days including native-speaker QA.

Where to go next

Dialect is where Indian elections are won and lost. The campaign that figures out the language stack in each state operates with a permanent edge over campaigns that don't.

Frequently asked questions

Why does dialect matter more in some states than others?

States with strong dialect identity (Rajasthan, Bihar, parts of UP and Maharashtra) see steep engagement drops when AI agents speak only standard Hindi. States with weaker dialect-identity politics (Delhi, urban Karnataka) tolerate standard Hindi/English better. The first group needs dialect tuning; the second doesn't.

How is Marwari different from Hindi for AI purposes?

Marwari uses different pronouns (थारो/म्हारो vs आपका/मेरा), different verb conjugations (जासी/होवेगा vs जाएगा/होगा), and a warmer set of family-related honorifics (काका/काकी). A Hindi-trained model can usually understand Marwari input but generates standard-Hindi responses unless explicitly prompted to use Marwari register.

Is Bihar one dialect or several?

Several. Bhojpuri dominates western Bihar (Champaran, Chapra, Siwan). Maithili dominates the north-east (Madhubani, Darbhanga). Magahi dominates south (Gaya, Nalanda, parts of Jharkhand border). Hindi dominates urban centres (Patna). A serious Bihar campaign needs all four configurations.

What's the Marathi dialect map?

Standard Marathi (Pune, Mumbai urban) dominates urban politics. Vidarbha Marathi has Hindi/Telugu influences and distinct vocabulary in eastern districts. Konkani-influenced coastal Marathi differs in Sindhudurg, Ratnagiri. Marathwada Marathi has historical Urdu/Persian influences. A Maharashtra-wide campaign needs 3-4 distinct configurations.

Should the candidate themselves speak the dialect?

Ideally yes for in-person events. For AI, the *agent* must speak the dialect even if the candidate doesn't. Voters are more forgiving of a candidate who speaks Hindi (it's the lingua franca) than of an AI that doesn't speak the local dialect (it signals inauthenticity).